Synergetic effects of Software and Hardware Parameters on the LSM system.
Talkat 12th USENIX Workshop on Hot Topics in Storage and File Systems (HotStorage '20) · 2020-07-13..07-14
summary
Shows that LSM tunables interact with storage type: fio benchmarks and RocksDB runs reveal HDD/SSD/NVMe/persistent memory behave differently — fio benchmark shows NVMe latency spikes under GC, mini‑SSD wins with high parallelism but breaks with large batch writes, persistent memory throughput sits far below theoretical limit and is sensitive to compaction scheduling. Introduces the quicksand effect hypothesis (higher media bandwidth shortens SSTable flushes, increasing compaction overhead) and a working device‑oriented tuning framework with a Python monitoring module and a RocksDB‑based data‑placement and migration prototype that moves data and formats between devices.
AI-written summary · about